verticals

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). Outsider free text is fetched at runtime from DuckDuckGo HTML (verticals/research.py::_fetch_ddg → r.text) and parsed into snippet strings that are inserted into the LLM prompt as “LIVE RESEARCH … untrusted raw text”.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill calls the DuckDuckGo HTML endpoint (https://html.duckduckgo.com/html/) at runtime in research_topic() and injects the returned snippets directly into the LLM prompt as live "RESEARCH" data, meaning an external URL's content is fetched during runtime and directly controls model instructions.